The article reports on a new study conducted using the Center for AI Safety (CAIS)–backed Remote Labor Index, which evaluated how well current AI agents perform in freelance-type tasks such as video editing, data scraping, and graphic design. The results are striking: even the top‐performing models completed less than 3% of the total simulated workload and earned only US$1,810 out of a possible $143,991.
A major reason for the dismal performance is that these agents lack key capabilities required for flexible, multi-step freelance work: long-term memory, the ability to learn on the job, and adaptiveness in ambiguous situations. Despite advances in coding, reasoning, and tool-use, current systems stumble outside narrow, well-defined tasks.
The study challenges the hype that autonomous agentic AI will soon replace human freelancers en‐masse. While generative models excel at discrete tasks (e.g., summarising text or generating first drafts), the messy, unbounded, context-rich nature of freelance gigs remains a significant barrier. The article suggests that for now, AI is a tool rather than a substitute for skilled human freelancers.
In closing, the piece argues organisations and freelancers should recalibrate their expectations: AI agents might assist or augment but are unlikely to fully take over freelance work in the short term. The human element — judgement, adaptability, context-awareness — remains critical. It’s a reminder that even as tools evolve, the structure of work and the skills required remain in flux.
